epic-array-simulation-benchmark

epic-array-simulation-benchmark

Summary

Validate DMR detection pipelines by reproducing expected DMR counts from the EPICSimData simulation dataset using ChAMP's bumphunter-based method. This skill benchmarks whether a methylation analysis workflow correctly identifies simulated differentially methylated regions and reveals detection thresholds (e.g., minimum CpG count per DMR).

When to use

When developing or validating a DNA methylation array analysis pipeline using ChAMP, you need an independent ground-truth dataset to confirm that DMR detection is working correctly. The EPICSimData simulation dataset is purpose-built for this: it contains synthetically injected DMRs with known locations and sizes, making it ideal for verifying that your champ.DMR() implementation produces the expected number of detected regions and behaves consistently with published benchmarks.

When NOT to use

• Input data is real patient or tissue samples, not a simulation — use this skill only for benchmarking; real data requires domain-specific interpretation and validation against biological phenotypes.
• You need to detect DMRs in a dataset where some regions legitimately contain only 1–2 CpGs — bumphunter's minimum CpG threshold may filter valid signal; consider alternative methods (Probe Lasso, DMRcate) if single-probe regions are biologically relevant.
• The goal is to identify cell-type-specific methylation patterns or infer cell composition — simulation benchmarking does not validate biological applicability in heterogeneous tissue samples.

Inputs

• EPICSimData methylation beta-value matrix (loaded via data(EPICSimData))
• ChAMP-compatible EPIC array annotation and sample metadata

Outputs

• DMR detection result object with list of identified regions
• DMR count (integer; expected ~4700+)
• DMR genomic coordinates and statistics (chr, start, end, p-value, etc.)

How to apply

Load the EPICSimData simulation dataset in R using data(EPICSimData), then execute the champ.DMR() function with the bumphunter-based detection method on the loaded methylation matrix. Extract and count the number of DMRs from the function output. Compare the observed count to the expected range (approximately 4700 or higher). If the count is significantly lower than expected, investigate whether the detection method is filtering out simulated DMRs that contain fewer than the minimum required CpGs (typically 2–3 CpGs per region in bumphunter). This reveals whether your normalization and preprocessing steps are preserving signal fidelity or whether detection thresholds need adjustment.